Diagnose AI agents that repeatedly call tools, revisit the same state, or fail to converge on a result.
Troubleshooting
An agent repeatedly calls the same tool or cycles without reaching a useful terminal state.
Add explicit termination and retry bounds and make tool results deterministic and actionable where possible.
Re-run the failing scenario and confirm the agent reaches a bounded terminal state.
Add loop detection, step budgets, traces, and regression scenarios.
Treat an agent as a state machine you can observe.
Trace:
goal → decision → tool call → tool result → state update → next decision
Look for repeated state, repeated tools, hidden errors, and missing terminal conditions.
Bound retries and steps, make tool failures explicit, and add loop-detection telemetry. Verify that the same scenario terminates predictably.
Engineering companion
Knowledge path